CN107025654A - The adaptive ship detection method of SAR image checked based on global iterative - Google Patents

The adaptive ship detection method of SAR image checked based on global iterative Download PDF

Info

Publication number
CN107025654A
CN107025654A CN201610070867.3A CN201610070867A CN107025654A CN 107025654 A CN107025654 A CN 107025654A CN 201610070867 A CN201610070867 A CN 201610070867A CN 107025654 A CN107025654 A CN 107025654A
Authority
CN
China
Prior art keywords
image
pixel
mask
ship
detection
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201610070867.3A
Other languages
Chinese (zh)
Other versions
CN107025654B (en
Inventor
田巳睿
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Science and Technology
Original Assignee
Nanjing University of Science and Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing University of Science and Technology filed Critical Nanjing University of Science and Technology
Priority to CN201610070867.3A priority Critical patent/CN107025654B/en
Publication of CN107025654A publication Critical patent/CN107025654A/en
Application granted granted Critical
Publication of CN107025654B publication Critical patent/CN107025654B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/88—Radar or analogous systems specially adapted for specific applications
    • G01S13/89—Radar or analogous systems specially adapted for specific applications for mapping or imaging
    • G01S13/90—Radar or analogous systems specially adapted for specific applications for mapping or imaging using synthetic aperture techniques, e.g. synthetic aperture radar [SAR] techniques
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/88—Radar or analogous systems specially adapted for specific applications
    • G01S13/89—Radar or analogous systems specially adapted for specific applications for mapping or imaging
    • G01S13/90—Radar or analogous systems specially adapted for specific applications for mapping or imaging using synthetic aperture techniques, e.g. synthetic aperture radar [SAR] techniques
    • G01S13/9021—SAR image post-processing techniques
    • G01S13/9029—SAR image post-processing techniques specially adapted for moving target detection within a single SAR image or within multiple SAR images taken at the same time
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/88—Radar or analogous systems specially adapted for specific applications
    • G01S13/89—Radar or analogous systems specially adapted for specific applications for mapping or imaging
    • G01S13/90—Radar or analogous systems specially adapted for specific applications for mapping or imaging using synthetic aperture techniques, e.g. synthetic aperture radar [SAR] techniques
    • G01S13/9021—SAR image post-processing techniques
    • G01S13/9027—Pattern recognition for feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10032—Satellite or aerial image; Remote sensing

Landscapes

  • Engineering & Computer Science (AREA)
  • Remote Sensing (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Physics & Mathematics (AREA)
  • Electromagnetism (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Radar Systems Or Details Thereof (AREA)
  • Image Processing (AREA)

Abstract

The invention discloses a kind of SAR (synthetic aperture radar) image adaptive ship detection method for being based on global iterative inspection (Global Iterative Censoring, GIC).This method extracts sea area using the geographic information data of imaging region；Heterogeneous sea is divided into multiple homogeneous subregions using the fuzzy C-mean algorithm grader (Fuzzy Local Information c Means clustering, FLICM) based on local location information；Using candidate's ship target in GIC technical mark all subregions, and the optimal sea clutter distributed model in synchronous self-adapting chosen area；The sea clutter model obtained according to GIC, is recognized using two-dimentional sliding window CFAR (CFAR) detector to candidate's ship target, improves accuracy of detection, realizes the detection to ship.Homogenous area partitioning technology and GIC are applied among SAR image ship detection by the present invention, have the advantages that sane efficient, high detection rate and low false alarm rate under the complex scenes such as heterogeneous sea clutter, multiple target scene.

Description

The adaptive ship detection method of SAR image checked based on global iterative
Technical field
It is specifically that one kind is examined based on region segmentation and global iterative the present invention relates to radar image processing technology field The SAR image sea ship self-adapting detecting method looked into.
Background technology
Ship target detection is the task with traditional of each littoral zone country of the world, is safeguarding national marine rights and interests, protection The many-sides such as marine environment, marine traffic control and Maritime Law Enforcement supervision have a wide range of applications.Synthetic aperture thunder Up to (Synthetic Aperture Radar, SAR) because of its round-the-clock, round-the-clock, a wide range of, high-resolution, length The advantage of phase Continuous Observation, has advantageous advantage, SAR image ship in terms of marine vessel detection Detection has also obtained extensive attention both domestic and external.In recent years, the development with SAR remote sensing technologies and a large amount of SAR The transmitting of satellite, carry out sea ship detection using SAR image has turned into the study hotspot that ocean remote sensing is applied.
At present, SAR image ship detection method is broadly divided into two classes, the i.e. detection to moving ship tail With the detection to ship body.
Wherein, to the detection of ship tail, the image procossings such as Hough transform, Radon conversion are mainly passed through Means, the surface wave tail for practicing midwifery raw by ship straight is searched in SAR image, by hull draining and spiral The turbulent wake that oar injection is caused, and the interior wave rear mark that ship is produced under the conditions of certain stratification.This kind of side The major defect of method is, does not simply fail to detect ship that is static or slowly moving, and motion ship Tail it is also and unstable, can usually be influenceed by the factor such as ship velocity and sea sea situation, in part SAR Even tail can not be observed on the image of sensor.The presence of tail can confirm the presence of ship, without Tail does not represent no ship.Therefore, current ship detection method is still based on the detection to ship body.
Detection to ship body mainly uses the scattering strength difference of ship and Sea background, i.e. ship-sea right Realized than degree difference.According to SAR imaging mechanisms, when being imaged to ship, hull and ocean surface, Hull partial structurtes can all constitute corner reflector, extremely strong back scattering be produced to radar wave, in SAR image Upper performance is the highlighted target being made up of several or even dozens of pixel.As the ocean surface of background, on sea When comparing tranquil (wind speed be less than 2m/s), sea shows as mirror-reflection to radar beam, backscattering echo compared with Weak, the marine background in SAR image is very dark；In sea wind than in the case of stronger, Bragg occurs for sea Resonance scattering, echo-signal is stronger, and the marine background in SAR image is partially bright.But in above-mentioned two situations Under, naval vessel backscattering echo is much stronger than marine background echo.Therefore warship is directly detected in SAR image Ship target, its essence is to detect highlighted target in it there is the dark background of clutter and noise jamming.In practical application In, influenceed by complicated marine environment and ship structure, there are a large amount of heterogeneous sea areas in SAR image, Medium-sized/small-sized vessel is influenceed by the sea clutter in heterogeneous region, and ship-sea contrast is weakened, detection performance by Influence.In addition, there is target-rich environment in the region such as harbour, navigation channel, marine fishing ground, a large amount of various sizes of ships Only it is gathered in zonule, part medium-sized/small-sized vessel is in detection by neighbouring large vessel echo and its secondary lobe Influence, detection parameter is elevated, and causes missing inspection, influences ship detection probability.
Existing SAR image ship detection method mainly includes following algorithm：Partitioning algorithm based on global threshold, Adaptive threshold method based on sliding window, the CFAR detection algorithms based on radar CFAR detection, based on template Threshold detection algorithm, and based on sub-aperture processing detection algorithm.
Wherein, global threshold algorithm is by setting a fixed global threshold to split image, by institute There is the pixel higher than set threshold value as Ship Target.It the advantage is that threshold calculations are simple, computational complexity is low, There is preferably detection performance to target-rich environment；Its shortcoming is that the algorithm can not become according to regional area in image Change and radar incidence angle changes adjust automatically threshold value, introduce a large amount of in the testing result of heterogeneous SAR image Missing inspection and false-alarm.
Adaptive threshold detection algorithm based on sliding window is directed to heterogeneous SAR image, utilizes window filtering technology Ship is detected, selected detection threshold value can preferably meet the statistical property of detection regional area. But this algorithm is made an uproar more in SAR image spot, when sea stormy waves is larger, or under target-rich environment, can be produced A large amount of false-alarms, influence detection performance.
CFAR (CFAR) detection method be while it is constant to ensure false alarm rate, according to false alarm rate and The statistical property (i.e. the probability density function of ocean clutter) of SAR image ocean clutter, which is calculated, to be obtained detecting naval vessel The threshold value of target.Its core concept is the selection of sea clutter model and the estimation of model parameter.According to for mould The sample of shape parameter estimation, CFAR detectors are further divided into global threshold CFAR detectors and two dimension is slided Window CFAR detectors.The former assumes whole sea area to be detected to meet the homogeneous area of given distributed model Domain, detection threshold value is calculated by the CFAR algorithms of standard.The latter then build by detection zone, warning region and The window of background area composition, it is assumed that the sea clutter in detection zone neighborhood meets given clutter mould Type, estimates Clutter Model parameter, and calculate detection threshold value by the sample in annular backdrop window.Due to detection Ship pixel quantity is far fewer than sea clutter background pixel quantity in region, and the Clutter Model used is generally long streaking Pattern type, global threshold CFAR detectors are influenceed smaller by target-rich environment, but to heterogeneous detection zone, This method will introduce substantial amounts of missing inspection and false-alarm in testing result.Compared with global threshold CFAR detectors, Sliding window CFAR detectors have more preferable robustness.But there is heterogeneous region in the reference zone of compound sliding window When border and target-rich environment, it detects that performance is even more worse than global threshold CFAR detectors.In addition, right Sea clutter is unsatisfactory for the situation of default clutter distributed model, global threshold CFAR detectors and sliding window CFAR The detection performance of detector will be all decreased obviously, and a large amount of false-alarms and missing inspection can be produced in testing result.
Detection algorithm based on template, is according to particular sensor SAR image and ship characteristic in region to be detected Priori, design one group of detection threshold value template detected.This method not only allows for naval vessel strong in itself Information is spent, while the strength information of its surrounding pixel also to be also served as detecting to a part for foundation, it is desirable to by drawing The information for entering surrounding pixel improves the accuracy of detection.Its major defect is that detection algorithm not only needs a large amount of numbers Priori is accumulated according to being observed with long-time, the data of particular sensor are also only applicable to, it is impossible to extensive use In the data of different sensors.In addition, this method equally can be by heterogeneous clutter zone boundary and multiple target ring The influence in border, introduces false-alarm and missing inspection in testing result.
Based on the detection algorithm of sub-aperture processing, by sub-band division, SAR complex patterns are decomposed into resolution ratio Relatively low multiple sub-aperture images, using ship and the sea clutter correlation between different pore size difference improve ship- Extra large contrast, can detect ship in the case of high sea situation, have certain suppression to heterogeneous SAR image Effect.But this method be to sacrifice resolution ratio as cost, it is therefore, closer to the distance under multiple target scene Ship may be mistaken for same target.In addition, medium and small ship is also due to resolution ratio reduction produces missing inspection.
In summary, existing all kinds of ship detection algorithms are still restricted by many factors, especially because radar imagery The heterogeneous SAR image influence that the factor such as incidence angle change and observation area sea situation is caused so that ship-sea contrast Degree declines, detection threshold value estimation mistake, occur in detection process missing inspection, false-alarm probability it is higher.In addition, right In intensive regions of ship such as harbour, navigation channel, marine fishing grounds, due to being mixed into ship in sea clutter background sample Sample, easily causes the estimation of clutter statistical parameter deviation occur, influences the threshold value of target detection, cause part ship Missing inspection, for the medium-sized/small-sized vessel in fishing ground and port area, this problem is particularly acute.
The content of the invention
, should the invention provides a kind of adaptive ship detection method of SAR image checked based on global iterative Method can not only improve missing inspection and false-alarm caused by target-rich environment, and homogeneous and heterogeneous SAR are schemed As being respectively provided with sane detectability and higher operation efficiency.
The technical solution for realizing the object of the invention is：A kind of SAR image checked based on global iterative is certainly Ship detection method is adapted to, is comprised the following steps：
Step 1：According to latitude and longitude coordinates by geographic information data it is registering with SAR image I after be superimposed, retain sea Pixel value in the domain of oceanic province, -1 is set to by the pixel value in other regions, obtains only retaining the SAR of sea area Image ImaskFor ship detection；
Step 2：Will by the numerical value of SAR image sea area pixel and position using LFICM partitioning algorithms ImaskIn sea area be divided into multiple homogeneous subregions, to avoid the zonule shadow produced in cutting procedure Ring, rejected the too small region of area using morphology closed operation；
Step 3：Global iterative inspection (GIC) and the selection of self-adapting clutter model are carried out by subregion；To each Subregion, iteration carry out based on AIC criterion optimal Clutter Model selection, global threshold CFAR detection and The ship target candidate image update of current region, until the ship target candidate image of current region does not change Become；Aforesaid operations are repeated to next region；All homogeneous subregions are traveled through, candidate's ship two-value is finally given Image；
Step 4：Each pixel in candidate's ship target bianry image is recognized, two-dimentional sliding window is utilized CFAR detectors calculate the corresponding local detection threshold value of current candidate ship pixel, if candidate's ship pixel value is small In detection threshold value, then it is set to background pixel in candidate's ship bianry image；Travel through all candidate pixels Afterwards, final ship detection result bianry image is obtained.
The segmentation of the SAR image homogenous area based on FLICM and Morphological scale-space described in step 2, by as follows Step is carried out：
2a) according to the sea situation characteristic of SAR image sea area, initial clustering quantity c, fuzzy degree factor are set M, iteration ends error ε and local neighborhood window size WFLICM；
The subordinated-degree matrix U of sea area all pixels 2b) is initialized using random number；Wherein, U=[uki]c×K (k=1 ..., c；I=1 ..., N) it is c × N matrix, N is ImaskMiddle sea area sum of all pixels, i.e. ImaskMiddle picture Plain value is not equal to -1 value；
2c) initialization calculates cluster centre：
Wherein, vk(k=1 ..., it is c) class center of k-th of cluster；xi(i=1 ..., N) is i-th in sea area Individual pixel；
2d) update the subordinated-degree matrix of sea area all pixels
Wherein, Gki(k=1 ..., c；I=1 ..., N) it is the x determined by local location informationiBelong to being subordinate to for k-th cluster The factor is spent, is calculated by following formula
Wherein, NiFor pixel xiW in SAR imageFLICM×WFLICMDeleted neighbourhood is (i.e. without center pixel WFLICM×WFLICMSquare window), xhFor h-th of pixel in the neighborhood；For xhAnd xiBetween Euclidean distance, (pi,qi) and (ph,qh) it is respectively xiAnd xhCoordinate in the picture；
2e) calculate subordinated-degree matrix change value Δ=| | Unew- U | |, wherein | | | | it is matrix 2-Norms computing； Iteration is terminated if Δ≤ε, 2f is performed)；Otherwise b=b+1, U=U are madenew, return to 2c) and continue iteration；
2f) set up and ImaskEtc. big image array Iseg, wherein all pixels value is set to -1；To pixel xi(pi,qi), make the homogeneous subregion classification belonging to itBy IsegMiddle coordinate For (pi,qi) pixel value be set to K；The all pixels of sea area are traveled through, that is, obtain the rough segmentation of sea area Cut image Iseg；
2g) to coarse segmentation image IsegSuccessively filtered and erosion filter using 5 × 5 morphological dilations, realization pair The suppression of zonule, so as to obtain the homogeneous sub-zone dividing image of sea areaIts InRepresent the union operation of c sub-regions, VKRepresentThe set that middle k-th subregion pixel is constituted, VlandForImage intermediate value is -1 pixel set, corresponding to ImaskThe set of Mid-continent domain pixel.
Subregion candidate's ship target detection based on GIC described in step 3 is selected with adaptive distributed model, Carry out as follows：
3a) definition and ImaskEtc. big two-value testing result imageWithWillWithIn all pictures Element is initialized as 0；Definition Clutter Model array M (k) corresponding with all subregion, k=1 ..., c；Set global Detect false-alarm probability
It is 3b) rightIn k-th subregion VK, extract the image coordinate of its all pixels
Wherein,It isMiddle coordinate is (pi,qi) pixel；According to coordinate set Scoord, extractIn The pixel value of correspondence position, obtains sets of pixel values
In ScoordEligible x is rejected in setdet(pi,qiThe coordinate sequence of)=1, the image coordinate after being checked SetMeet
According to setExtract ImaskThe pixel of middle correspondence position, builds model sample estimates collectionMeet
Wherein, MKFor setIn number of elements；
3c) using AIC criterion from Gaussian Profile, rayleigh distributed, Gamma are distributed, K is distributed, Weibull Distribution and G0 are distributed model of the selection with minimum entropy in six conventional distributions and portray the extra large miscellaneous of current sub-region Wavelength-division cloth；Will setMiddle pixel value is divided into ξ grade, uses yj, j=1 ..., ξ characterize j-th of grade, And have y1＜ y2＜ ... ＜ yξ, then definable AIC verify as
Wherein, ∏ () multiplies symbol to connect；fModel(yj) it is Gaussian Profile, rayleigh distributed, Gamma distributions, K Distribution, Weibull distributions and G0 are distributed in six conventional distributions, the clutter probability density levied by Model table Function, OFPFor the number of parameters of the probability-density function of clutter；
The Optimal Distribution model for causing the model of E (Model) value minimum as current region is chosen, and records mould The estimate of shape parameter, for being detected to target in region；
3d) subregion V is calculated using CFAR algorithmsKGlobal detection threshold value
Wherein, T is the subregion V obtained using golden section search algorithmKGlobal detection threshold value；
3e) by detection threshold value T and subregion VKInterior all pixels value is compared, and is more than T's for pixel value Situation, willThe pixel value of correspondence position is set to 1 in image；
3f) compareWithThe difference of image, if two images have differences, makesWillIn Subregion VKInterior pixel value is set to 0, repeat step 3b-3e, untilWithIndifference, and by current region Clutter Model charge in Clutter Model array k-th element M (K)；
3b-3f steps 3g) are repeated to next subregion, until traversalIn all homogeneous subregions；It is defeated Go out GIC testing result bianry imagesWith all subregion Clutter Model array M (k), k=1 ..., c.
The identification of candidate's ship pixel based on two-dimentional sliding window CFAR detectors described in step 4, as follows Carry out：
4a) according to ship size and region characteristic in detection marine site, two-dimentional sliding window CFAR detectors are set respectively Reference windows size Rref, alert window size RG, it is single pixel, definition identification to set detection window size False alarm rate isWith minimum number of reference nref；The structure of two-dimentional sliding window CFAR detectors is as shown in Figure 1.
Two-dimentional sliding window CFAR detectors 4b) are applied to ImaskImage；
To ImaskPixel x in imagemask(p0,q0), the detection unit of sliding window CFAR detectors is placed in xmask(p0,q0) position；Then, ring protection regionWith annular reference regionIt is covered in xmask(p0,q0) Rref×RrefIn neighborhood；If it is in bianry imageMiddle correspondence The pixel x of positiondet(p0,q0)=1, then extract ImaskIt is located at x in imagemask(p0,q0) in neighborhood reference zone Pixel xmask(pj,qj), whereinExist respectivelyWithMiddle extraction correspondence position pixel value xdet(pj,qj) and xseg(pj,qj), set up reference zone set of pixels Close Sref
Sref={ (xmask(pj,qj),xdet(pj,qj),xseg(pj,qj))} (10)
In set, take and meet xdet(pj,qjThe set element of)=0 sets up model parameter estimation sample set Sest
Sest={ (xmask(pj,qj),xdet(pj,qj),xseg(pj,qj))|xdet(pj,qj)=0 } (11)
According to xseg(pj,qj) correspondence homogenous area, by SestIt is divided into ClocalIndividual subset (Clocal≤c)
Wherein, Ci∈ [1, c] is subsetThe corresponding homogeneous subarea number of middle element, i=1 ..., Clocal.OrderCharacterize subsetMiddle number of elements,Represent subsetMiddle xmask(pj,qj) it is equal Value, then with maximum xmask(pj,qj) average subsetIt is defined as
IfThen useClutter Model M (the C of correspondence homogeneous subregionmax), and profit With the x in setmask(pj,qj) element estimation Clutter Model parameter, obtain the corresponding probability-density function of clutterOtherwise, using center pixel region C0Clutter Model M (C0), utilize SestIn All samples estimate Clutter Model parameter, obtain the corresponding probability-density function of clutter
Judge its corresponding false-alarm probability using following formula
IfThen willX in imagedet(p0,q0) it is set to 0；
Above-mentioned steps 4c) are pressed, I is traveled throughmaskThe all pixels of image, after identificationImage is exported, i.e., It can obtain final detection result bianry image
It is of the invention compared with existing SAR ship detections technology, its remarkable advantage is：
(1) the invention provides to homogeneous and heterogeneous SAR image and harbour, navigation channel, marine fishing ground etc. Target-rich environment provides a kind of sane reliable detection method, and high detection probability is respectively provided with all cases With relatively low false-alarm probability, and detection Robust Performance in all cases；
(2) the invention provides an adaptively selected strategy to heterogeneous SAR image Clutter Model, keep away Exempt from due to presetting the detection hydraulic performance decline that single Clutter Model is caused；
(3) it the composite can be widely applied to the SAR of different sensors, different working modes and different resolution Data, with good universality；
(4) present invention has the characteristics of computation complexity is low, processing speed is fast, can be to the magnanimity SAR of big breadth Data carry out quick ship detection process.
Brief description of the drawings
Fig. 1 is two-dimentional sliding window CFAR detector structural representations.
Fig. 2 is adaptive ship detection algorithm overview flow chart.
Fig. 3 is that homogeneous sea area divides flow chart.
Fig. 4 is that sub-region iteration checks process chart.
Fig. 5 is the sub-region candidate target identification process chart based on optimal sea clutter distributed model.
Fig. 6 is experiment SAR image used.
Fig. 7 is the result that Fig. 6 extracts sea area by geographic information data.
Fig. 8 is homogenous area division result.
Fig. 9 is red boxes inner region area image in Fig. 7, and the present invention, two-dimentional sliding window K-CFAR detectors and Comparison of the OMW K-CFAR detection methods to its testing result.
Embodiment
Accompanying drawings below is described in further detail to the present invention.
A kind of adaptive ship detection method of SAR image checked based on global iterative, by based on FLICM's Adaptive homogenous area division, the adaptive subregion Clutter Model selection based on GIC and ship coarse segmentation, And three key steps such as the ship target identification based on two-dimentional sliding window constant false alarm detector are constituted.
Adaptive homogenous area based on FLICM is divided, first by geographic information data by sea to be detected Oceanic province domain is split from the SAR image of input, is divided into sea area using FLICM algorithms some Individual homogeneous subregion, and the zonule closed using morphology in filtering rejecting segmentation result.The processing final output Image is divided for homogenous area
The selection of adaptive subregion Clutter Model and ship coarse segmentation based on GIC, by traveling through area to be detected Every sub-regions in domain realize the selection of self-adapting clutter model and ship coarse segmentation to all subregion.To pending Current sub-region, the present invention selects optimal miscellaneous in one's respective area by GIC methods first with AIC criterion Ripple distributed model, recycles optimal Clutter Model to calculate the global CFAR detection threshold values of current region, obtains and works as The ship coarse segmentation result of forefoot area, then will be judged as the sample data of ship pixel from the back of the body of current region Aforesaid operations are repeated after being rejected in scape sample.Until no longer changing in the detection of current region.Traversal is treated Every sub-regions of detection zone, final output coarse segmentation result imagesWith Clutter Model array M (K).
Based on the ship target identification of two-dimentional sliding window CFAR detectors, using two-dimentional Slide-window detector to coarse segmentation The testing result marked in result images is recognized.It is right in SAR imageLabeled as each picture of target The two-dimentional sliding window constant false alarm detector of element application, to avoid the interference of target-rich environment, will be labeled in reference zone Rejected for the sample of ship pixel, the homogenous area dividing condition according to present in background area chooses suitably miscellaneous Wave pattern and background sample estimation parameter, calculate identification threshold value, realize identification.It is rightIn it is used labeled After pixel identification, final output testing result bianry image
The present invention uses workflow as shown in Figure 2：
Step 1：According to latitude and longitude coordinates by geographic information data it is registering with SAR image I after be superimposed, retain sea Pixel value in the domain of oceanic province, -1 is set to by the pixel value in other regions, obtains only retaining the SAR of sea area Image ImaskFor ship detection.
Step 2：Using flow as shown in Figure 3, by ImaskIn sea area be divided into multiple homogeneous subregions.
2a) according to the sea situation characteristic of SAR image sea area, initial clustering quantity c, fuzzy degree factor are set M, iteration ends error ε and local neighborhood window size WFLICM。
The subordinated-degree matrix U of sea area all pixels 2b) is initialized using random number.Wherein, U=[uki]c×K (k=1 ..., c；I=1 ..., N) it is c × N matrix, N is ImaskMiddle sea area sum of all pixels, i.e. ImaskMiddle picture Plain value is not equal to -1 value.
2c) initialization calculates cluster centre：
Wherein, vk(k=1 ..., it is c) class center of k-th of cluster；xi(i=1 ..., N) is i-th in sea area Individual pixel.
2d) update the subordinated-degree matrix of sea area all pixels
Wherein, Gki(k=1 ..., c；I=1 ..., N) it is the x determined by local location informationiBelong to being subordinate to for k-th cluster The factor is spent, can be calculated by following formula
Wherein, NiFor pixel xiW in SAR imageFLICM×WFLICMDeleted neighbourhood is (i.e. without center pixel WFLICM×WFLICMSquare window), xhFor h-th of pixel in the neighborhood；For xhAnd xiBetween Euclidean distance, (pi,qi) and (ph,qh) it is respectively xiAnd xhCoordinate in the picture.
2e) calculate subordinated-degree matrix change value Δ=| | Unew- U | |, wherein | | | | it is matrix 2-Norms computing. Iteration is terminated if Δ≤ε, 2f is performed)；Otherwise b=b+1, U=U are madenew, return to 2c) and continue iteration.
2f) set up and ImaskEtc. big image array Iseg, wherein all pixels value is set to -1.To pixel xi(pi,qi), make the homogeneous subregion classification belonging to itBy IsegMiddle coordinate For (pi,qi) pixel value be set to K.Travel through all pixels of sea area, you can obtain the thick of sea area Segmentation figure is as Iseg。
2g) to coarse segmentation image IsegSuccessively filtered and erosion filter using 5 × 5 morphological dilations, realization pair The suppression of zonule, so as to obtain the homogeneous sub-zone dividing image of sea areaIts InRepresent the union operation of c sub-regions, VKRepresentThe set that middle k-th subregion pixel is constituted, VlandForImage intermediate value is -1 pixel set, corresponding to ImaskThe set of Mid-continent domain pixel.
Step 3：Using flow as shown in Figure 4, global iterative inspection (GIC) is carried out and adaptive by subregion Clutter Model is selected.
3a) set up and ImaskEtc. big two-value testing result imageWithWillWithMiddle institute There is pixel to be initialized as 0.
3b) basisMiddle homogeneous sub-zone dividing situation, initialization sub-area processing variable K=0, manually The false-alarm probability of overall situation CFAR detections is setSet up the corresponding Clutter Model array of all subregion M (k), k=1 ..., c.
3c) existIn, extract k-th subregion VKThe image coordinate of all pixels
Wherein,It isMiddle coordinate is (pi,qi) pixel.According to coordinate set Scoord, extractIn The pixel value of correspondence position, obtains sets of pixel values
In ScoordEligible x is rejected in setdet(pi,qiThe coordinate sequence of)=1, the image coordinate after being checked SetMeet
According to setExtract ImaskThe pixel of middle correspondence position, builds model sample estimates collectionMeet
Wherein, MKFor setIn number of elements.
3d) using AIC criterion from Gaussian Profile, rayleigh distributed, Gamma are distributed, K is distributed, Weibull Distribution and G0 are distributed model of the selection with minimum entropy in six conventional distributions and portray the extra large miscellaneous of current sub-region Wavelength-division cloth.Will setMiddle pixel value is divided into ξ grade, uses yj, j=1 ..., ξ characterize j-th of grade, And have y1＜ y2＜ ... ＜ yξ, then definable AIC verify as
Wherein, ∏ () multiplies symbol to connect；fModel(yj) be distributed for Gaussian Profile, rayleigh distributed, Gamma, K distributions, Weibull distributions and G0 are distributed in six conventional distributions, the clutter probability density levied by Model table Function, OFPFor the number of parameters of the probability-density function of clutter.
The probability density function of Gaussian Profile may be defined as
Wherein, μ is setMiddle pixel average；σ2For setMiddle pixel variance.
The probability density function of rayleigh distributed is defined as
Wherein,
The probability density function of Gamma distributions is defined as
Wherein, L is that radar regards number, is provided by radar head fileinfo；Γ () is Gamma functions, and v is shape The factor, can be tried to achieve by following formula
Wherein, Γ ' (x) is Γ () first derivative.
The probability density function of K distributions is defined as
Wherein, γ is form factor, is solved and obtained by following formula：
Kγ-L() is γ-L rank first kind modified Bessel functions.
The probability density function of Weibull distributions is defined as
Wherein, b scale parameters, a is form parameter, can be tried to achieve by following formula
The probability density function of G0 distributions is defined as
Wherein, α and lambda parameter can be solved by following formula：
The Optimal Distribution model for causing the model of E (Model) value minimum as current region is chosen, and records mould The estimate of shape parameter, for being detected to target in region.
3e) subregion V is calculated using CFAR algorithmKGlobal detection threshold value
Wherein, T is the subregion V obtained using golden section search algorithmKGlobal detection threshold value.
3f) by detection threshold value T and subregion VKInterior all pixels value is compared, and is more than T's for pixel value Situation, willThe pixel value of correspondence position is set to 1 in image.
3g) compareWithThe difference of image, if two images have differences, makesWillIn Subregion VKRespective pixel value is set to 0, repeat step 3c-3f, untilWithIndifference, and proparea will be worked as The Clutter Model in domain is charged in Clutter Model array k-th element M (K).
3c-3g steps 3h) are repeated to next subregion, until traversalIn all homogeneous subregions.It is defeated Go out GIC testing result bianry imagesWith all subregion Clutter Model array M (k), k=1 ..., c.
Step 4：Using flow as shown in Figure 5, using two-dimentional sliding window CFAR detectors to candidate's ship target Recognized.
4a) according to ship size and region characteristic in detection marine site, two-dimentional sliding window CFAR detectors are set respectively Reference windows size Rref, alert window size RG, it is single pixel, definition identification to set detection window size False alarm rate isWith minimum number of reference nref.The structure of two-dimentional sliding window CFAR detectors is as shown in Figure 1.
The two-dimentional sliding window CFAR detectors shown in Fig. 1 4b) are applied to ImaskImage.
To ImaskPixel x in imagemask(p0,q0), the detection unit of sliding window CFAR detectors is placed in xmask(p0,q0) position.Then, the ring protection region shown in Fig. 1With annular ginseng Examination district domainIt is covered in xmask(p0,q0) Rref×RrefIn neighborhood.If it is in bianry image The pixel x of middle correspondence positiondet(p0,q0)=1, then extract ImaskIt is located at x in imagemask(p0,q0) neighborhood reference Pixel x in regionmask(pj,qj), wherein Exist respectivelyWithMiddle extraction correspondence position pixel value xdet(pj,qj) and xseg(pj,qj), set up reference zone Pixel set Sref
Sref={ (xmask(pj,qj),xdet(pj,qj),xseg(pj,qj))} (37)
In set, take and meet xdet(pj,qjThe set element of)=0 sets up model parameter estimation sample set Sest
Sest={ (xmask(pj,qj),xdet(pj,qj),xseg(pj,qj))|xdet(pj,qj)=0 } (38)
According to xseg(pj,qj) correspondence homogenous area, by SestIt is divided into ClocalIndividual subset (Clocal≤c)
Wherein, Ci∈ [1, c] is subsetThe corresponding homogeneous subarea number of middle element, i=1 ..., Clocal.OrderCharacterize subsetMiddle number of elements,Represent subsetMiddle xmask(pj,qj) it is equal Value, then with maximum xmask(pj,qj) average subsetIt may be defined as
IfThen useClutter Model M (the C of correspondence homogeneous subregionmax), and profit With the x in setmask(pj,qj) element estimation Clutter Model parameter, obtain the corresponding probability-density function of clutterOtherwise, using center pixel region C0Clutter Model M (C0), utilize SestIn All samples estimate Clutter Model parameter, obtain the corresponding probability-density function of clutter
Judge its corresponding false-alarm probability using following formula
IfThen willX in imagedet(p0,q0) it is set to 0.
Above-mentioned steps 4c) are pressed, I is traveled throughmaskThe all pixels of image, after identificationImage is exported, i.e., It can obtain final testing result bianry image
With reference to embodiment, the present invention will be further described.
Embodiment 1
The present invention is a kind of adaptive ship detection method of SAR image checked based on global iterative, is mainly related to And the detection to ship target on the region sea such as harbour, navigation channel, fishing ground, specific implementation is by shown in Fig. 2 What step was realized：
Step 1：Water area SAR image based on geographic information data is extracted.
To the SAR image I to be detected of input, using the latitude and longitude coordinates in header file, SAR is schemed As I is superimposed with the extra large land Registration of Measuring Data in geographic information database.Retain the pixel value in sea area, by it SAR image pixel value in its region is set to -1, obtain only retaining the SAR image I of sea areamask。
Step 2：According to the handling process shown in Fig. 3, using FLICM algorithms and morphologic filtering, by Imask Sea area in image is divided into several homogeneous subregions.
2a) according to input sea area SAR image Imask={ xi| i=1 ..., N } sea situation characteristic, manually set Initial clustering quantity c, fuzzy degree factor m, iteration ends error ε and local neighborhood window size WFLICM。
C × N-dimensional subordinated-degree matrix U=[u of sea area all pixels 2b) are initialized using random numberki]c×K。
Cluster centre v 2c) is calculated using formula (16) according to subordinated-degree matrix Uk(k=1 ..., c).
2d) according to cluster centre vk(k=1 ..., c), is calculated by W using formula (17)FLICM×WFLICMGo the heart adjacent The degree of membership factor matrix G=[G that local location information is determined in domainki]c×N, then update sea using formula (17) The subordinated-degree matrix of oceanic province domain all pixels
2e) calculate subordinated-degree matrix change value Δ=| | Unew-U||.Iteration is terminated if Δ≤ε, 2f is performed)； Otherwise b=b+1, U=U are madenew, return to 2c) and continue iteration.
2f) set up and ImaskEtc. big image array Iseg, wherein all pixels value is set to -1.To pixel xi(pi,qi), make the homogeneous subregion classification belonging to itBy IsegMiddle coordinate For (pi,qi) pixel value be set to K.Travel through all pixels of sea area, you can obtain the thick of sea area Segmentation figure is as Iseg。
2g) to coarse segmentation image IsegSuccessively filtered and erosion filter using 5 × 5 morphological dilations, realization pair The suppression of zonule, so as to obtain the homogeneous sub-zone dividing image of sea areaIts InRepresent the union operation of c sub-regions, VKRepresentThe set that middle k-th subregion pixel is constituted, VlandForImage intermediate value is -1 pixel set, corresponding to ImaskThe set of Mid-continent domain pixel.
Step 4：According to handling process described in Fig. 4, global iterative inspection (GIC) is carried out and adaptive by subregion Clutter Model is answered to select.
3a) set up and ImaskEtc. big two-value testing result imageWithBy all pixels of two images It is initialized as 0.
3b) basisMiddle homogeneous sub-zone dividing situation, initialization sub-area processing variable K=0, manually The false-alarm probability of overall situation CFAR detections is setSet up the corresponding Clutter Model array of all subregion M (k), k=1 ..., c.
3c) according to formula (19)-(22), I is extractedmaskMiddle pixel, builds model sample estimates collection
3d) according to formula (23)-(35), divided using AIC criterion from Gaussian Profile, rayleigh distributed, Gamma Cloth, K distributions, Weibull distributions and G0 are distributed model of the selection with minimum entropy in six conventional distributions and carved Draw the sea clutter distribution of current sub-region.
3e) according to formula (36), using golden section search algorithm, subregion V is obtainedKGlobal detection threshold value T。
3f) by detection threshold value T and subregion VKInterior all pixels value is compared, and is more than T's for pixel value Situation, willThe pixel value of correspondence position is set to 1 in image.
3g) compareWithThe difference of image, if two images have differences, makesWillIn Subregion VKRespective pixel value is set to 0, repeat step 3b-3e, untilWithIndifference, and proparea will be worked as The Clutter Model in domain is charged in the element M (K) of Clutter Model array.
3b-3f steps 3h) are repeated to next subregion, until traversalIn all homogeneous subregions.It is defeated Go out GIC testing result bianry imagesWith all subregion Clutter Model array M (k), k=1 ..., c.
Step 4：Using flow as shown in Figure 5, using two-dimentional sliding window CFAR detectors to candidate's ship target Recognized.
4a) according to ship size and region characteristic in detection marine site, two-dimentional sliding window CFAR detectors are set respectively Reference windows size Rref, alert window size RG, it is single pixel, definition identification to set detection window size False alarm rate isWith minimum number of reference nref.The structure of two-dimentional sliding window CFAR detectors is as shown in Figure 1.
4b) to ImaskPixel x in imagemask(p0,q0), the detection unit of Slide-window detector is placed in xmask(p0,q0) position.Then, the ring protection region shown in Fig. 1With annular ginseng Examination district domainIt is covered in xmask(p0,q0) Rref×RrefIn neighborhood.If it is in bianry image The pixel x of middle correspondence positiondet(p0,q0)=1, then according to formula (37)-(38), set up model parameter estimation sample Set Sest。
4c) according to SestThe difference of middle element correspondence homogenous area, by SestIt is divided into ClocalIndividual subset (Clocal≤ c), i.e.,Wherein, Ci∈ [1, c] is subsetThe corresponding homogeneous sub-district of middle element Field Number, i=1 ..., Clocal.OrderCharacterize subsetMiddle number of elements,Represent son CollectionMiddle xmask(pj,qj) average,For with maximum xmask(pj,qj) average subset.
If 4d)Then useClutter Model M (the C of correspondence homogeneous subregionmax), And utilize the x in setmask(pj,qj) element estimation Clutter Model parameter, obtain corresponding clutter probability density FunctionOtherwise, using center pixel region C0Clutter Model M (C0), utilize Sest In all samples estimation Clutter Model parameters, obtain the corresponding probability-density function of clutter
4e) the corresponding local false-alarm probability of center pixel is calculated using formula (42)
If 4f)Then willX in imagedet(p0,q0) it is set to 0；Otherwise do not process.
If 4g) current pixel is not ImaskIn last pixel, then by two-dimentional sliding window CFAR detectors move To ImaskIn next pixel, return to step 4b).
4h) by after identificationImage is assigned toAnd export, you can obtain final detection result binary map Picture
Embodiment 2
The adaptive ship detection method be the same as Example 1 of SAR image checked based on global iterative, it is of the invention Parameter setting and effect are further illustrated by the experiment below to measured data：
Test the measured data used smart for the C-band Radarsat-1 single polarizations obtained on January 7th, 2000 Thin mode data, data distance to azimuth resolution as 6.25 meters.The map of magnitudes of measured data is such as Shown in Fig. 6, the world coastline data that the geographic information data used provides for European Space Agency are tested.
Experiment parameter sets as follows：For the homogenous area division processing based on FLICM, initial clustering is set Quantity c=6, fuzzy degree factor m=2, iteration ends error ε=1 × 10-6With local neighborhood window size WFLICM=5.Handled for global iterative inspection (GIC) and the selection of self-adapting clutter model, overall situation CFAR is set The false-alarm probability of detectionDistinguished for candidate's ship based on two-dimentional sliding window CFAR detectors Knowledge is handled, and sets the reference windows size R of two-dimentional sliding window CFAR detectorsref=75, alert window size RG=53, it is single pixel to set detection window size, and definition identification false alarm rate isWith minimum ginseng Examine unit number nref=280.
Sea area extracts result images and sees Fig. 7, and homogenous area division result is shown in Fig. 8.For ease of showing and comparing Compared with the Detection results of the inventive method, the regional area that rectangular block is indicated in Fig. 7 is taken to illustrate, the region Enlarged drawing see Fig. 9 (a).Fig. 9 (b) is to take adding for the inventive method to survey result；Fig. 9 (c) is using identical The two-dimentional sliding window K-CFAR detector processes results of window parameter；Fig. 9 (d) is to be set using identical parameters OMW K-CFAR detection algorithm results, it can be seen that the algorithm introduces substantial amounts of in testing result False-alarm.Obviously, compared to this method, other conventional methods will introduce more false-alarm.
In summary, the invention provides a kind of adaptive ship detection method of sane SAR image, for Existing detection algorithm detected under heterogeneous sea SAR image and target-rich environment hydraulic performance decline and robustness compared with Poor the problem of, use automatic homogenous area to divide and select clutter type and detect performance to solve heterogeneous region Influence, solved by the way of global iterative inspection and two-dimentional sliding window CFAR detections are combined target-rich environment with The influence that homogeneous subzone boundaries are estimated detection threshold value.The present invention significantly improves the ship under complicated marine environment Detectability, available for the ship target in detection homogeneous and heterogeneous sea SAR image.

Claims (4)

1. a kind of adaptive ship detection method of SAR image checked based on global iterative, it is characterised in that bag Containing following steps：
Step 1：According to latitude and longitude coordinates by geographic information data it is registering with SAR image I after be superimposed, retain sea Pixel value in the domain of oceanic province, -1 is set to by the pixel value in other regions, obtains only retaining the SAR of sea area Image ImaskFor ship detection；
Step 2：Will by the numerical value of SAR image sea area pixel and position using LFICM partitioning algorithms ImaskIn sea area be divided into multiple homogeneous subregions, using morphology closed operation by the too small region of area Reject；
Step 3：Global iterative inspection (GIC) and the selection of self-adapting clutter model are carried out by subregion；To each Subregion, iteration carry out based on AIC criterion optimal Clutter Model selection, global threshold CFAR detection and The ship target candidate image update of current region, until the ship target candidate image of current region does not change Become；Aforesaid operations are repeated to next region；All homogeneous subregions are traveled through, candidate's ship two-value is finally given Image；
Step 4：Each pixel in candidate's ship target bianry image is recognized, two-dimentional sliding window is utilized CFAR detectors calculate the corresponding local detection threshold value of current candidate ship pixel, if candidate's ship pixel value is small In detection threshold value, then it is set to background pixel in candidate's ship bianry image；Travel through all candidate pixels Afterwards, final ship detection result bianry image is obtained.
2. the SAR image adaptive ship detection side according to claim 1 checked based on global iterative Method, it is characterised in that at the segmentation of the SAR image homogenous area based on FLICM and morphology described in step 2 Reason, is carried out as follows：
2a) according to the sea situation characteristic of SAR image sea area, initial clustering quantity c, fuzzy degree factor are set M, iteration ends error ε and local neighborhood window size WFLICM；
The subordinated-degree matrix U of sea area all pixels 2b) is initialized using random number；Wherein, U=[uki]c×K (k=1 ..., c；I=1 ..., N) it is c × N matrix, N is ImaskMiddle sea area sum of all pixels, i.e. ImaskMiddle picture Plain value is not equal to -1 value；
2c) initialization calculates cluster centre：
Wherein, vk(k=1 ..., it is c) class center of k-th of cluster；xi(i=1 ..., N) is i-th in sea area Individual pixel；
2d) update the subordinated-degree matrix of sea area all pixels
Wherein, Gki(k=1 ..., c；I=1 ..., N) it is the x determined by local location informationiBelong to being subordinate to for k-th cluster The factor is spent, is calculated by following formula
Wherein, NiFor pixel xiW in SAR imageFLICM×WFLICMDeleted neighbourhood is (i.e. without center pixel WFLICM×WFLICMSquare window), xhFor h-th of pixel in the neighborhood；For xhAnd xiBetween Euclidean distance, (pi,qi) and (ph,qh) it is respectively xiAnd xhCoordinate in the picture；
2e) calculate subordinated-degree matrix change value Δ=| | Unew- U | |, wherein | | | | it is matrix 2-Norms computing； Iteration is terminated if Δ≤ε, 2f is performed)；Otherwise b=b+1, U=U are madenew, return to 2c) and continue iteration；
2f) set up and ImaskEtc. big image array Iseg, wherein all pixels value is set to -1；To pixel xi(pi,qi), make the homogeneous subregion classification belonging to itBy IsegMiddle coordinate For (pi,qi) pixel value be set to K；The all pixels of sea area are traveled through, that is, obtain the rough segmentation of sea area Cut image Iseg；
2g) to coarse segmentation image IsegSuccessively filtered and erosion filter using 5 × 5 morphological dilations, realization pair The suppression of zonule, so as to obtain the homogeneous sub-zone dividing image of sea areaIts InRepresent the union operation of c sub-regions, VKRepresentThe set that middle k-th subregion pixel is constituted, VlandForImage intermediate value is -1 pixel set, corresponding to ImaskThe set of Mid-continent domain pixel.
3. the SAR image adaptive ship detection side according to claim 1 checked based on global iterative Method, it is characterised in that subregion candidate's ship target detection based on GIC described in step 3 is divided with adaptive Cloth model is selected, and is carried out as follows：
3a) definition and ImaskEtc. big two-value testing result imageWithWillWithIn all pictures Element is initialized as 0；Definition Clutter Model array M (k) corresponding with all subregion, k=1 ..., c；Set global Detect false-alarm probability
It is 3b) rightIn k-th subregion VK, extract the image coordinate of its all pixels
Wherein,It isMiddle coordinate is (pi,qi) pixel；According to coordinate set Scoord, extractIn The pixel value of correspondence position, obtains sets of pixel values
In ScoordEligible x is rejected in setdet(pi,qiThe coordinate sequence of)=1, the image coordinate after being checked SetMeet
According to setExtract ImaskThe pixel of middle correspondence position, builds model sample estimates collectionMeet
Wherein, MKFor setIn number of elements；
3c) using AIC criterion from Gaussian Profile, rayleigh distributed, Gamma are distributed, K is distributed, Weibull Distribution and G0 are distributed model of the selection with minimum entropy in six conventional distributions and portray the extra large miscellaneous of current sub-region Wavelength-division cloth；Will setMiddle pixel value is divided into ξ grade, uses yj, j=1 ..., ξ characterize j-th of grade, And have y1＜ y2＜ ... ＜ yξ, then definable AIC verify as
Wherein, Π () multiplies symbol to connect；fModel(yj) it is Gaussian Profile, rayleigh distributed, Gamma distributions, K Distribution, Weibull distributions and G0 are distributed in six conventional distributions, the clutter probability density levied by Model table Function, OFPFor the number of parameters of the probability-density function of clutter；
The Optimal Distribution model for causing the model of E (Model) value minimum as current region is chosen, and records mould The estimate of shape parameter, for being detected to target in region；
3d) subregion V is calculated using CFAR algorithmsKGlobal detection threshold value
Wherein, T is the subregion V obtained using golden section search algorithmKGlobal detection threshold value；
3e) by detection threshold value T and subregion VKInterior all pixels value is compared, and is more than T's for pixel value Situation, willThe pixel value of correspondence position is set to 1 in image；
3f) compareWithThe difference of image, if two images have differences, makesWillIn Subregion VKInterior pixel value is set to 0, repeat step 3b-3e, untilWithIndifference, and by current region Clutter Model charge in Clutter Model array k-th element M (K)；
3b-3f steps 3g) are repeated to next subregion, until traversalIn all homogeneous subregions；It is defeated Go out GIC testing result bianry imagesWith all subregion Clutter Model array M (k), k=1 ..., c.
4. the SAR image adaptive ship detection side according to claim 1 checked based on global iterative Method, it is characterised in that the identification of candidate's ship pixel based on two-dimentional sliding window CFAR detectors described in step 4, Carry out as follows：
4a) according to ship size and region characteristic in detection marine site, two-dimentional sliding window CFAR detectors are set respectively Reference windows size Rref, alert window size RG, it is single pixel, definition identification to set detection window size False alarm rate isWith minimum number of reference nref；
Two-dimentional sliding window CFAR detectors 4b) are applied to ImaskImage；
To ImaskPixel x in imagemask(p0,q0), the detection unit of sliding window CFAR detectors is placed in xmask(p0,q0) position；Then, ring protection regionWith annular reference regionIt is covered in xmask(p0,q0) Rref×RrefIn neighborhood；If it is in bianry imageMiddle correspondence The pixel x of positiondet(p0,q0)=1, then extract ImaskIt is located at x in imagemask(p0,q0) in neighborhood reference zone Pixel xmask(pj,qj), wherein maskmask00mask00mask00refrefdet00maskmask00maskjjDetjjsegjjref exists respectivelyWithMiddle extraction correspondence position pixel value xdet(pj,qj) and xseg(pj,qj), set up reference zone set of pixels Close Sref
Sref={ (xmask(pj,qj),xdet(pj,qj),xseg(pj,qj))} (10)
In set, take and meet xdet(pj,qjThe set element of)=0 sets up model parameter estimation sample set Sest
Sest={ (xmask(pj,qj),xdet(pj,qj),xseg(pj,qj))|xdet(pj,qj)=0 } (11)
According to xseg(pj,qj) correspondence homogenous area, by SestIt is divided into ClocalIndividual subset (Clocal≤c)
(13)
xdet(pj,qj)=0, xseg(pj,qj)=Ci}
Wherein, Ci∈ [1, c] is subsetThe corresponding homogeneous subarea number of middle element, i=1 ..., Clocal.OrderCharacterize subsetMiddle number of elements,Represent subsetMiddle xmask(pj,qj) it is equal Value, then with maximum xmask(pj,qj) average subsetIt is defined as
IfThen useClutter Model M (the C of correspondence homogeneous subregionmax), and profit With the x in setmask(pj,qj) element estimation Clutter Model parameter, obtain the corresponding probability-density function of clutterOtherwise, using center pixel region C0Clutter Model M (C0), utilize SestIn All samples estimate Clutter Model parameter, obtain the corresponding probability-density function of clutter
Judge its corresponding false-alarm probability using following formula
IfThen willX in imagedet(p0,q0) it is set to 0；
Above-mentioned steps 4c) are pressed, I is traveled throughmaskThe all pixels of image, after identificationImage is exported, i.e., It can obtain final detection result bianry image
CN201610070867.3A 2016-02-01 2016-02-01 SAR image self-adaptive ship detection method based on global iterative inspection Active CN107025654B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610070867.3A CN107025654B (en) 2016-02-01 2016-02-01 SAR image self-adaptive ship detection method based on global iterative inspection

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610070867.3A CN107025654B (en) 2016-02-01 2016-02-01 SAR image self-adaptive ship detection method based on global iterative inspection

Publications (2)

Publication Number Publication Date
CN107025654A true CN107025654A (en) 2017-08-08
CN107025654B CN107025654B (en) 2020-07-31

Family

ID=59524475

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610070867.3A Active CN107025654B (en) 2016-02-01 2016-02-01 SAR image self-adaptive ship detection method based on global iterative inspection

Country Status (1)

Country Link
CN (1) CN107025654B (en)

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107589409A (en) * 2017-08-22 2018-01-16 电子科技大学 One kind splits antenna MIMO radar distribution low traffic detection fusion method
CN108957421A (en) * 2018-08-03 2018-12-07 中国人民解放军战略支援部队信息工程大学 One kind is based on Bezier fitting Radar emitter individual discrimination method and system
CN109001725A (en) * 2018-06-07 2018-12-14 中国人民解放军海军工程大学 A kind of sea unmanned boat sea multi-object tracking method
CN109358330A (en) * 2018-11-06 2019-02-19 电子科技大学 A kind of biradical Forward-looking SAR linear session synchronous error estimation method
CN109584279A (en) * 2017-09-29 2019-04-05 中国科学院遥感与数字地球研究所 Ship detecting method, apparatus and ship detecting system based on SAR image
CN109766811A (en) * 2018-12-31 2019-05-17 复旦大学 The end-to-end detection and recognition methods of sea ship in a kind of satellite-borne SAR image
CN110426681A (en) * 2019-07-31 2019-11-08 长春理工大学 A kind of LFM modulated parameter estimating method based on synchronous extraction S-transformation
CN112986991A (en) * 2021-02-05 2021-06-18 中国人民解放军国防科技大学 Clutter distribution independent polarimetric synthetic aperture radar ship detection method
CN113256667A (en) * 2021-04-14 2021-08-13 大连海事大学 SAR image ship target detection method based on Otsu and joint distribution
CN113466866A (en) * 2021-08-23 2021-10-01 西北工业大学 Moving target high-resolution synthetic aperture radar imaging algorithm based on tracking
CN113936199A (en) * 2021-12-17 2022-01-14 珠海视熙科技有限公司 Image target detection method and device and camera equipment
CN114463365A (en) * 2022-04-12 2022-05-10 中国空气动力研究与发展中心计算空气动力研究所 Infrared weak and small target segmentation method, device and medium

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
MAOGUO GONG等: "Change Detection in Synthetic Aperture Radar Images based on Image Fusion and Fuzzy Clustering", 《IEEE》 *
WENTAO AN 等: "An Improved Iterative Censoring Scheme for CFAR Ship Detection With SAR Imagery", 《IEEE》 *
YI CUI 等: "On the Iterative Censoring for Target Detection in SAR Images", 《IEEE》 *
田巳睿 等: "基于引力场增强的SAR图像舰船检测方法研究", 《遥感学报》 *
谭昆: "SAR图像舰船目标检测方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *
赵明波 等: "SAR图像CFAR检测的快速算法综述", 《自动化学报》 *

Cited By (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107589409A (en) * 2017-08-22 2018-01-16 电子科技大学 One kind splits antenna MIMO radar distribution low traffic detection fusion method
CN107589409B (en) * 2017-08-22 2020-07-21 电子科技大学 Distributed low-traffic detection fusion method for MIMO radar with separate antennas
CN109584279A (en) * 2017-09-29 2019-04-05 中国科学院遥感与数字地球研究所 Ship detecting method, apparatus and ship detecting system based on SAR image
CN109001725A (en) * 2018-06-07 2018-12-14 中国人民解放军海军工程大学 A kind of sea unmanned boat sea multi-object tracking method
CN109001725B (en) * 2018-06-07 2020-11-10 中国人民解放军海军工程大学 Offshore unmanned ship offshore multi-target tracking method
CN108957421B (en) * 2018-08-03 2020-09-18 中国人民解放军战略支援部队信息工程大学 Radar radiation source individual identification method and system based on Bezier curve fitting
CN108957421A (en) * 2018-08-03 2018-12-07 中国人民解放军战略支援部队信息工程大学 One kind is based on Bezier fitting Radar emitter individual discrimination method and system
CN109358330A (en) * 2018-11-06 2019-02-19 电子科技大学 A kind of biradical Forward-looking SAR linear session synchronous error estimation method
CN109358330B (en) * 2018-11-06 2020-04-21 电子科技大学 Bistatic forward-looking SAR linear time synchronization error estimation method
CN109766811A (en) * 2018-12-31 2019-05-17 复旦大学 The end-to-end detection and recognition methods of sea ship in a kind of satellite-borne SAR image
CN109766811B (en) * 2018-12-31 2023-05-02 复旦大学 End-to-end detection and identification method for sea surface ship in spaceborne SAR image
CN110426681A (en) * 2019-07-31 2019-11-08 长春理工大学 A kind of LFM modulated parameter estimating method based on synchronous extraction S-transformation
CN110426681B (en) * 2019-07-31 2023-07-14 长春理工大学 LFM signal parameter estimation method based on synchronous extraction S transformation
CN112986991B (en) * 2021-02-05 2022-06-21 中国人民解放军国防科技大学 Clutter distribution independent polarimetric synthetic aperture radar ship detection method
CN112986991A (en) * 2021-02-05 2021-06-18 中国人民解放军国防科技大学 Clutter distribution independent polarimetric synthetic aperture radar ship detection method
CN113256667A (en) * 2021-04-14 2021-08-13 大连海事大学 SAR image ship target detection method based on Otsu and joint distribution
CN113256667B (en) * 2021-04-14 2023-07-21 大连海事大学 SAR image ship target detection method based on Otsu and joint distribution
CN113466866A (en) * 2021-08-23 2021-10-01 西北工业大学 Moving target high-resolution synthetic aperture radar imaging algorithm based on tracking
CN113466866B (en) * 2021-08-23 2024-06-07 西北工业大学 Moving target high-resolution synthetic aperture radar imaging algorithm based on tracking
CN113936199B (en) * 2021-12-17 2022-05-13 珠海视熙科技有限公司 Image target detection method and device and camera equipment
CN113936199A (en) * 2021-12-17 2022-01-14 珠海视熙科技有限公司 Image target detection method and device and camera equipment
CN114463365B (en) * 2022-04-12 2022-06-24 中国空气动力研究与发展中心计算空气动力研究所 Infrared weak and small target segmentation method, equipment and medium
CN114463365A (en) * 2022-04-12 2022-05-10 中国空气动力研究与发展中心计算空气动力研究所 Infrared weak and small target segmentation method, device and medium

Also Published As

Publication number Publication date
CN107025654B (en) 2020-07-31

Similar Documents

Publication Publication Date Title
CN107025654A (en) The adaptive ship detection method of SAR image checked based on global iterative
Hou et al. FUSAR-Ship: Building a high-resolution SAR-AIS matchup dataset of Gaofen-3 for ship detection and recognition
CN107145874B (en) Ship target detection and identification method in complex background SAR image
CN109766811B (en) End-to-end detection and identification method for sea surface ship in spaceborne SAR image
Wei et al. HRSID: A high-resolution SAR images dataset for ship detection and instance segmentation
Zhao et al. A coupled convolutional neural network for small and densely clustered ship detection in SAR images
CN109583369B (en) Target identification method and device based on target area segmentation network
Chen et al. Ship detection from coastal surveillance videos via an ensemble Canny-Gaussian-morphology framework
CN110378308B (en) Improved port SAR image near-shore ship detection method based on fast R-CNN
Solberg et al. Automatic detection of oil spills in Envisat, Radarsat and ERS SAR images
Wang et al. An improved faster R-CNN based on MSER decision criterion for SAR image ship detection in harbor
Ding et al. Performance analysis and validation of waterline extraction approaches using single-and dual-polarimetric SAR data
CN109117802A (en) Ship Detection towards large scene high score remote sensing image
CN106709914A (en) SAR image ship detection false alarm eliminating method based on two-stage DEM sea-land reservoir
Hou et al. End-to-end automatic ship detection and recognition in high-resolution Gaofen-3 spaceborne SAR images
Pastina et al. Detection of ship targets in COSMO-SkyMed SAR images
Ding et al. Coastline detection in SAR images using multiscale normalized cut segmentation
Chen et al. Shape similarity intersection-over-union loss hybrid model for detection of synthetic aperture radar small ship objects in complex scenes
Song et al. An efficient marine oil spillage identification scheme based on an improved active contour model using fully polarimetric SAR imagery
Chen et al. SAR ship target detection for SSDv2 under complex backgrounds
CN112101250A (en) Method for detecting offshore ship target based on context semantic perception
Tarmizi et al. Shoreline data extraction from QuickBird satellite image using semi-automatic technique
Yang et al. A novel ship detection method based on sea state analysis from optical imagery
Wang et al. An inshore ship detection method in SAR images based on contextual fluctuation information
CN108764016B (en) Polarimetric SAR image ship detection method based on rotation domain characteristics and CNN

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant